Save Audio (Silent)

ComfyUI treats "output nodes" specially: they mark the end of the graph and only execute when the whole prompt finishes. That's exactly right for a final SaveImage. It's a nightmare inside a loop. When you want audio saved on every iteration - a TTS batch, per-scene audio in a video pipeline, one music clip per prompt - a terminal output node tends not to fire per-pass, so you end up with one file or zero and no obvious reason why. Looping in ComfyUI is fiddly enough without that.

Save Audio (Silent) exists for precisely that situation. It writes a WAV to your ComfyUI output folder, but it is deliberately not an output node - so it executes inline, and it passes the audio through unchanged on its own audio output. Chain the loop's audio through it and you get a file written every iteration instead of a skipped one.

Inputs: audio (AUDIO), filename_prefix (default "audio"), and an optional index INT. Output: audio - the same audio, so the graph keeps flowing.

A few honest notes about what it does and doesn't do:

• Filenames get the ComfyUI auto-increment treatment, same as SaveImage: audio_00001_.wav, audio_00002_.wav… The optional index input is accepted but doesn't actually change the path - the counter is generated for you. Ignore it.
• It saves via torchaudio, imported lazily inside the function. If your install somehow lacks torchaudio it'll fail at save time with an import error, but modern ComfyUI ships it, so this is rarely the problem.
• It copes with the several audio-dict conventions floating around the ecosystem (waveform/samples/data, sample_rate/sr/sampling_rate), falling back to 44100 Hz if none are present. More forgiving than most save nodes bother to be.
• If the audio input is None it prints a warning and skips the save rather than crashing - handy on partial or cancelled batches.

Install is the same as its packmates - one clone gets you all three nodes:

cd ComfyUI/custom_nodes
git clone https://github.com/kenyonxu/ComfyUI_KaiUtilities.git

Restart ComfyUI and it's under KaiUtilities/audio → Save Audio (Silent). Or grab it through ComfyUI Manager by searching the pack title "Kai Civitai LoRA Loader."

The name is honest: "silent" in the sense that it produces no terminal behavior and makes no fuss - it saves and gets out of the way. The stock SaveAudio is fine for a single generation; the moment you wrap generation in a loop, swap in this one. Small, single-purpose, and exactly right for the one workflow pattern it targets.